[AIX] Implement the ifunc attribute. #153049

w2yehia · 2025-08-11T16:43:52Z

Currently, the AIX linker and loader do not provide a mechanism to implement ifuncs similar to GNU_ifunc on ELF Linux.
On AIX, we will lower __attribute__((ifunc("resolver")) to the llvm ifunc as other platforms do. The llvm ifunc in turn will get lowered at late stages of the optimization pipeline to an AIX-specific implementation. No special linkage or relocations are needed when generating assembly/object output.

On AIX, a function foo has two symbols associated with it: a function descriptor (foo) residing in the .data section, and an entry point (.foo) residing in the .text section. The first field of the descriptor is the address of the entry point. Typically, the address field in the descriptor is initialized once: statically, at load time (?), or at runtime if runtime linking is enabled.

Here we would like to use the address field in the descriptor to implement the ifunc semantics. Specifically, the ifunc function will become a stub that jumps to the entry point in the address field. A constructor function is linked into every linkage module. The constructor walks an array of {descriptor, resolver} pairs, calling the resolver and saving the result in the address field in the descriptor.

TODO:

Add tests for:
- some permutations of
  1. 32/64-bit
  2. successful link with and without ifuncs
  3. bcdtors:mbr and bcdtors:all
  4. -ffunction-sections and -fno-function-sections
  5. -fno-integrated-as
- define foo in a shared object and (1) call it from main, (2) call from another shared object
- define foo as weak, static, hidden, exported, protected
- alias to an ifunc

llvmbot · 2025-08-11T16:44:23Z

@llvm/pr-subscribers-clang-driver
@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-backend-powerpc

Author: Wael Yehia (w2yehia)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/153049.diff

10 Files Affected:

(modified) clang/include/clang/Basic/TargetInfo.h (+2)
(modified) llvm/include/llvm/IR/IntrinsicsPowerPC.td (+10)
(modified) llvm/include/llvm/Transforms/Utils/ModuleUtils.h (+7)
(modified) llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp (+6-2)
(modified) llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp (+4)
(modified) llvm/lib/Target/PowerPC/PPCISelLowering.cpp (+26)
(modified) llvm/lib/Target/PowerPC/PPCTargetMachine.cpp (+11)
(modified) llvm/lib/Target/PowerPC/PPCTargetMachine.h (+2)
(modified) llvm/lib/Transforms/Utils/LowerIFunc.cpp (+6-1)
(modified) llvm/lib/Transforms/Utils/ModuleUtils.cpp (+81)

diff --git a/clang/include/clang/Basic/TargetInfo.h b/clang/include/clang/Basic/TargetInfo.h
index ce4677e540226..ffeab27b67911 100644
--- a/clang/include/clang/Basic/TargetInfo.h
+++ b/clang/include/clang/Basic/TargetInfo.h
@@ -1545,6 +1545,8 @@ class TargetInfo : public TransferrableTargetInfo,
       return true;
     if (getTriple().getArch() == llvm::Triple::ArchType::avr)
       return true;
+    if (getTriple().isOSAIX())
+      return true;
     return getTriple().isOSBinFormatELF() &&
            ((getTriple().isOSLinux() && !getTriple().isMusl()) ||
             getTriple().isOSFreeBSD());
diff --git a/llvm/include/llvm/IR/IntrinsicsPowerPC.td b/llvm/include/llvm/IR/IntrinsicsPowerPC.td
index 7dd9ff7f08b8b..75b64fba6e3ca 100644
--- a/llvm/include/llvm/IR/IntrinsicsPowerPC.td
+++ b/llvm/include/llvm/IR/IntrinsicsPowerPC.td
@@ -2090,3 +2090,13 @@ let TargetPrefix = "ppc" in {
     Intrinsic<[], [llvm_i64_ty, llvm_i64_ty, llvm_ptr_ty],
               [IntrArgMemOnly, IntrWriteMem, NoCapture<ArgIndex<2>>]>;
 }
+
+
+//===----------------------------------------------------------------------===//
+// XCOFF Intrinsics
+let TargetPrefix = "ppc" in {
+  def int_ppc_get_function_entry :
+      DefaultAttrsIntrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
+  def int_ppc_get_function_descriptor :
+      DefaultAttrsIntrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
+}
diff --git a/llvm/include/llvm/Transforms/Utils/ModuleUtils.h b/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
index 4036c4e947c75..6be7ac80dd428 100644
--- a/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
+++ b/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
@@ -159,6 +159,13 @@ LLVM_ABI bool
 lowerGlobalIFuncUsersAsGlobalCtor(Module &M,
                                   ArrayRef<GlobalIFunc *> IFuncsToLower = {});
 
+/// AIX specific lowering of ifuncs where we convert an ifunc to a regular
+/// function with the following implementation:
+///  Check if the function's descriptor still points to itself (true on first
+/// entry), if so then call the resolver function and atomically store the 
+/// resulting function pointer into the descriptor. Make an indirect call 
+/// through the function pointer in the descriptor.
+LLVM_ABI void lowerIFuncsOnAIX(Module &M);
 } // End llvm namespace
 
 #endif // LLVM_TRANSFORMS_UTILS_MODULEUTILS_H
diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 2ab2c147be0ec..957b2b3e7555d 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -763,6 +763,8 @@ static MCSymbol *getMCSymbolForTOCPseudoMO(const MachineOperand &MO,
     return AP.GetJTISymbol(MO.getIndex());
   case MachineOperand::MO_BlockAddress:
     return AP.GetBlockAddressSymbol(MO.getBlockAddress());
+  case MachineOperand::MO_MCSymbol:
+     return MO.getMCSymbol();
   default:
     llvm_unreachable("Unexpected operand type to get symbol.");
   }
@@ -792,6 +794,8 @@ getTOCEntryTypeForMO(const MachineOperand &MO) {
     return PPCAsmPrinter::TOCType_JumpTable;
   case MachineOperand::MO_BlockAddress:
     return PPCAsmPrinter::TOCType_BlockAddress;
+  case MachineOperand::MO_MCSymbol:
+    return PPCAsmPrinter::TOCType_GlobalExternal; // TODO
   default:
     llvm_unreachable("Unexpected operand type to get TOC type.");
   }
@@ -1043,7 +1047,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
     TmpInst.setOpcode(PPC::LWZ);
 
     const MachineOperand &MO = MI->getOperand(1);
-    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress()) &&
+    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress() || MO.isMCSymbol()) &&
            "Invalid operand for LWZtoc.");
 
     // Map the operand to its corresponding MCSymbol.
@@ -1127,7 +1131,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
     TmpInst.setOpcode(PPC::LD);
 
     const MachineOperand &MO = MI->getOperand(1);
-    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress()) &&
+    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress() || MO.isMCSymbol()) &&
            "Invalid operand!");
 
     // Map the operand to its corresponding MCSymbol.
diff --git a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 415164fc9e2cb..921cb72a83030 100644
--- a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -6158,6 +6158,10 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
         replaceWith(PPC::ADDItoc8, N, MVT::i64);
         return;
       }
+      if (N->getOperand(0).getOpcode() == ISD::MCSymbol) {
+        replaceWith(PPC::LDtoc, N, MVT::i64);
+        return;
+      }
       // Break if it doesn't have toc data attribute. Proceed with common
       // SelectCode.
       break;
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 74ae8502dccea..bbdd573f4a81b 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -11167,6 +11167,32 @@ SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
   SDLoc dl(Op);
 
   switch (IntrinsicID) {
+  case Intrinsic::ppc_get_function_descriptor:
+    return Op.getOperand(1);
+  case Intrinsic::ppc_get_function_entry: {
+    SDValue Op1 = Op.getOperand(1);
+    if (auto *G = dyn_cast<GlobalAddressSDNode>(Op1)) {
+      const GlobalValue *GV = G->getGlobal();
+      assert(isFunctionGlobalAddress(GV));
+      assert(Subtarget.isAIXABI());
+      assert(!isa<GlobalIFunc>(GV) && "IFunc is not supported on AIX.");
+      const TargetMachine &TM = Subtarget.getTargetMachine();
+      const TargetLoweringObjectFile *TLOF = TM.getObjFileLowering();
+      MCSymbolXCOFF *S =
+          cast<MCSymbolXCOFF>(TLOF->getFunctionEntryPointSymbol(GV, TM));
+
+      MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
+      SDValue EntryPointSym = DAG.getMCSymbol(S, PtrVT);
+      return getTOCEntry(DAG, dl, EntryPointSym);
+    }
+    assert (Op1.getOpcode() == ISD::CopyFromReg);
+    SDLoc dl(Op);
+    EVT PtrVT = getPointerTy(DAG.getDataLayout());
+    SDValue Result = DAG.getLoad(
+            PtrVT, dl, DAG.getEntryNode(), Op1, MachinePointerInfo(), DAG.getDataLayout().getPointerABIAlignment(0),
+            MachineMemOperand::MODereferenceable);
+    return Result;
+  }
   case Intrinsic::thread_pointer:
     // Reads the thread pointer register, used for __builtin_thread_pointer.
     if (Subtarget.isPPC64())
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
index b5c6ac111dff0..d8e840912bc37 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -36,6 +36,7 @@
 #include "llvm/InitializePasses.h"
 #include "llvm/MC/TargetRegistry.h"
 #include "llvm/Pass.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
@@ -43,6 +44,7 @@
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/TargetParser/Triple.h"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/LowerIFunc.h"
 #include <cassert>
 #include <memory>
 #include <optional>
@@ -452,6 +454,15 @@ class PPCPassConfig : public TargetPassConfig {
 
 } // end anonymous namespace
 
+void PPCTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
+  if (getTargetTriple().isOSAIX())
+    PB.registerOptimizerLastEPCallback(
+        [](ModulePassManager &PM, OptimizationLevel Level, ThinOrFullLTOPhase Phase) {
+          if (Phase == ThinOrFullLTOPhase::None)
+            PM.addPass(LowerIFuncPass());
+        });
+}
+
 TargetPassConfig *PPCTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new PPCPassConfig(*this, PM);
 }
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.h b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
index cb02b446fadb3..608750f7a3f2b 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
@@ -54,6 +54,8 @@ class PPCTargetMachine final : public CodeGenTargetMachineImpl {
   // Pass Pipeline Configuration
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
 
+  void registerPassBuilderCallbacks(PassBuilder &PB) override;
+
   TargetTransformInfo getTargetTransformInfo(const Function &F) const override;
 
   TargetLoweringObjectFile *getObjFileLowering() const override {
diff --git a/llvm/lib/Transforms/Utils/LowerIFunc.cpp b/llvm/lib/Transforms/Utils/LowerIFunc.cpp
index 18ae0bbe2e731..991890aa56b83 100644
--- a/llvm/lib/Transforms/Utils/LowerIFunc.cpp
+++ b/llvm/lib/Transforms/Utils/LowerIFunc.cpp
@@ -13,6 +13,7 @@
 #include "llvm/Transforms/Utils/LowerIFunc.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
+#include "llvm/TargetParser/Triple.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
 
 using namespace llvm;
@@ -22,6 +23,10 @@ PreservedAnalyses LowerIFuncPass::run(Module &M, ModuleAnalysisManager &AM) {
   if (M.ifunc_empty())
     return PreservedAnalyses::all();
 
-  lowerGlobalIFuncUsersAsGlobalCtor(M, {});
+  Triple TargetTriple(M.getTargetTriple());
+  if (TargetTriple.isOSAIX())
+    lowerIFuncsOnAIX(M);
+  else
+    lowerGlobalIFuncUsersAsGlobalCtor(M, {});
   return PreservedAnalyses::none();
 }
diff --git a/llvm/lib/Transforms/Utils/ModuleUtils.cpp b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
index 596849ecab742..67da3e66cf16a 100644
--- a/llvm/lib/Transforms/Utils/ModuleUtils.cpp
+++ b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
@@ -16,6 +16,8 @@
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsPowerPC.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/Casting.h"
@@ -397,6 +399,85 @@ void llvm::embedBufferInModule(Module &M, MemoryBufferRef Buf,
   appendToCompilerUsed(M, GV);
 }
 
+void llvm::lowerIFuncsOnAIX(Module &M) {
+  for (GlobalIFunc &IFunc : make_early_inc_range(M.ifuncs())) {
+
+    // Let IFunc be:
+    //   @foo = ifunc rt-type (type1, type2), ptr @resolver
+    assert(isa<llvm::FunctionType>(IFunc.getValueType()));
+    FunctionType *FTy = cast<llvm::FunctionType>(IFunc.getValueType());
+
+    // Create the regular (non-ifunc) function:
+    //   define rt-type @foo(type1 %0, type2 %1)
+    Function *F = Function::Create(FTy, IFunc.getLinkage(), IFunc.getAddressSpace(), IFunc.getName(), &M);
+    LLVMContext &Ctx = F->getContext();
+
+    // entry:
+    BasicBlock *CurBlock = BasicBlock::Create(Ctx, "entry", F);
+    IRBuilder<> Builder(CurBlock);
+
+    PointerType *PtrTy = Builder.getPtrTy();
+    //   %DescPtr = call ptr @ppc_get_function_descriptor(ptr noundef @foo)
+    auto *DescPtr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_descriptor, {F}, {}, "desc.ptr");
+
+    //   %DesctAddr = getelementptr inbounds %struct.Desc_t, ptr %DescPtr, i32 0, i32 0
+    StructType *DescriptorType = StructType::get(PtrTy, PtrTy, PtrTy);
+    auto *DesctAddr = Builder.CreateStructGEP(DescriptorType, DescPtr, 0, "desc_t.addr");
+
+    //   %AddrInDesc = load ptr, ptr %DesctAddr, align 4
+    auto *AddrInDesc = Builder.CreateAlignedLoad(PtrTy, DesctAddr, DesctAddr->getPointerAlignment(M.getDataLayout()), "addr.in.desc");
+
+    //   %OriginalAddr = call ptr @ppc_get_function_entry(ptr noundef @foo) #3
+    auto *OriginalAddr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_entry, {F}, {}, "original.addr");
+
+    //   %cmp = icmp eq ptr %AddrInDesc, %OriginalAddr
+    auto *CMP = Builder.CreateICmpEQ(AddrInDesc, OriginalAddr);
+
+    //   br i1 %cmp, label %resolver, label %end
+    CurBlock = BasicBlock::Create(Ctx, "resolver", F);
+    BasicBlock *FinalBlock = BasicBlock::Create(Ctx, "end", F);
+    Builder.CreateCondBr(CMP, CurBlock, FinalBlock);
+
+    // Emit the 'then' (resolver) code:
+    Builder.SetInsertPoint(CurBlock);
+
+    // resolver:
+    //   %ResolvedFunc = call ptr @resolver()
+    //   %ResolvedAddr = call ptr @ppc_get_function_entry(ptr %ResolvedFunc)
+    //   %4 = ptrtoint ptr %ResolvedAddr to i32
+    //   store atomic i32 %4, ptr %desc_t.addr release, align 4
+    Function *ResolverFunc = IFunc.getResolverFunction();
+    auto *ResolvedFunc = Builder.CreateCall(ResolverFunc, ArrayRef<Value *>(), "resolved.func");
+    auto *ResolvedAddr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_entry, {ResolvedFunc}, {}, "resolved.addr");
+    auto *PtrToInt = Builder.CreatePtrToInt(ResolvedAddr, Builder.getIntPtrTy(M.getDataLayout()));
+    // TODO fix alignment
+    Builder.CreateAlignedStore(PtrToInt, DesctAddr, MaybeAlign())->setAtomic(AtomicOrdering::Release);
+
+    //   br label %if.end
+    Builder.CreateBr(FinalBlock);
+
+    // Emit the continuation block for code after the if.
+    Builder.SetInsertPoint(FinalBlock);
+
+    //   %res = musttail call i32 %DescPtr(i32 noundef %a) #3
+    SmallVector<Value *, 10> Args(make_pointer_range(F->args()));
+    CallInst *Result = Builder.CreateCall(F->getFunctionType(), DescPtr, Args, "res");
+    //Result->setTailCallKind(CallInst::TCK_MustTail);
+
+    //   ret i32 %res
+    if (F->getReturnType()->isVoidTy())
+      Builder.CreateRetVoid();
+    else
+      Builder.CreateRet(Result);
+
+    // replace all uses of the ifunc with the newly created function
+    IFunc.replaceAllUsesWith(F);
+
+    std::string name = IFunc.getName().str();
+    IFunc.eraseFromParent();
+    F->setName(name);
+  }
+}
 bool llvm::lowerGlobalIFuncUsersAsGlobalCtor(
     Module &M, ArrayRef<GlobalIFunc *> FilteredIFuncsToLower) {
   SmallVector<GlobalIFunc *, 32> AllIFuncs;

llvmbot · 2025-08-11T16:44:24Z

@llvm/pr-subscribers-llvm-ir

Author: Wael Yehia (w2yehia)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/153049.diff

10 Files Affected:

(modified) clang/include/clang/Basic/TargetInfo.h (+2)
(modified) llvm/include/llvm/IR/IntrinsicsPowerPC.td (+10)
(modified) llvm/include/llvm/Transforms/Utils/ModuleUtils.h (+7)
(modified) llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp (+6-2)
(modified) llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp (+4)
(modified) llvm/lib/Target/PowerPC/PPCISelLowering.cpp (+26)
(modified) llvm/lib/Target/PowerPC/PPCTargetMachine.cpp (+11)
(modified) llvm/lib/Target/PowerPC/PPCTargetMachine.h (+2)
(modified) llvm/lib/Transforms/Utils/LowerIFunc.cpp (+6-1)
(modified) llvm/lib/Transforms/Utils/ModuleUtils.cpp (+81)

diff --git a/clang/include/clang/Basic/TargetInfo.h b/clang/include/clang/Basic/TargetInfo.h
index ce4677e540226..ffeab27b67911 100644
--- a/clang/include/clang/Basic/TargetInfo.h
+++ b/clang/include/clang/Basic/TargetInfo.h
@@ -1545,6 +1545,8 @@ class TargetInfo : public TransferrableTargetInfo,
       return true;
     if (getTriple().getArch() == llvm::Triple::ArchType::avr)
       return true;
+    if (getTriple().isOSAIX())
+      return true;
     return getTriple().isOSBinFormatELF() &&
            ((getTriple().isOSLinux() && !getTriple().isMusl()) ||
             getTriple().isOSFreeBSD());
diff --git a/llvm/include/llvm/IR/IntrinsicsPowerPC.td b/llvm/include/llvm/IR/IntrinsicsPowerPC.td
index 7dd9ff7f08b8b..75b64fba6e3ca 100644
--- a/llvm/include/llvm/IR/IntrinsicsPowerPC.td
+++ b/llvm/include/llvm/IR/IntrinsicsPowerPC.td
@@ -2090,3 +2090,13 @@ let TargetPrefix = "ppc" in {
     Intrinsic<[], [llvm_i64_ty, llvm_i64_ty, llvm_ptr_ty],
               [IntrArgMemOnly, IntrWriteMem, NoCapture<ArgIndex<2>>]>;
 }
+
+
+//===----------------------------------------------------------------------===//
+// XCOFF Intrinsics
+let TargetPrefix = "ppc" in {
+  def int_ppc_get_function_entry :
+      DefaultAttrsIntrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
+  def int_ppc_get_function_descriptor :
+      DefaultAttrsIntrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
+}
diff --git a/llvm/include/llvm/Transforms/Utils/ModuleUtils.h b/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
index 4036c4e947c75..6be7ac80dd428 100644
--- a/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
+++ b/llvm/include/llvm/Transforms/Utils/ModuleUtils.h
@@ -159,6 +159,13 @@ LLVM_ABI bool
 lowerGlobalIFuncUsersAsGlobalCtor(Module &M,
                                   ArrayRef<GlobalIFunc *> IFuncsToLower = {});
 
+/// AIX specific lowering of ifuncs where we convert an ifunc to a regular
+/// function with the following implementation:
+///  Check if the function's descriptor still points to itself (true on first
+/// entry), if so then call the resolver function and atomically store the 
+/// resulting function pointer into the descriptor. Make an indirect call 
+/// through the function pointer in the descriptor.
+LLVM_ABI void lowerIFuncsOnAIX(Module &M);
 } // End llvm namespace
 
 #endif // LLVM_TRANSFORMS_UTILS_MODULEUTILS_H
diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 2ab2c147be0ec..957b2b3e7555d 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -763,6 +763,8 @@ static MCSymbol *getMCSymbolForTOCPseudoMO(const MachineOperand &MO,
     return AP.GetJTISymbol(MO.getIndex());
   case MachineOperand::MO_BlockAddress:
     return AP.GetBlockAddressSymbol(MO.getBlockAddress());
+  case MachineOperand::MO_MCSymbol:
+     return MO.getMCSymbol();
   default:
     llvm_unreachable("Unexpected operand type to get symbol.");
   }
@@ -792,6 +794,8 @@ getTOCEntryTypeForMO(const MachineOperand &MO) {
     return PPCAsmPrinter::TOCType_JumpTable;
   case MachineOperand::MO_BlockAddress:
     return PPCAsmPrinter::TOCType_BlockAddress;
+  case MachineOperand::MO_MCSymbol:
+    return PPCAsmPrinter::TOCType_GlobalExternal; // TODO
   default:
     llvm_unreachable("Unexpected operand type to get TOC type.");
   }
@@ -1043,7 +1047,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
     TmpInst.setOpcode(PPC::LWZ);
 
     const MachineOperand &MO = MI->getOperand(1);
-    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress()) &&
+    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress() || MO.isMCSymbol()) &&
            "Invalid operand for LWZtoc.");
 
     // Map the operand to its corresponding MCSymbol.
@@ -1127,7 +1131,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
     TmpInst.setOpcode(PPC::LD);
 
     const MachineOperand &MO = MI->getOperand(1);
-    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress()) &&
+    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI() || MO.isBlockAddress() || MO.isMCSymbol()) &&
            "Invalid operand!");
 
     // Map the operand to its corresponding MCSymbol.
diff --git a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 415164fc9e2cb..921cb72a83030 100644
--- a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -6158,6 +6158,10 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
         replaceWith(PPC::ADDItoc8, N, MVT::i64);
         return;
       }
+      if (N->getOperand(0).getOpcode() == ISD::MCSymbol) {
+        replaceWith(PPC::LDtoc, N, MVT::i64);
+        return;
+      }
       // Break if it doesn't have toc data attribute. Proceed with common
       // SelectCode.
       break;
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 74ae8502dccea..bbdd573f4a81b 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -11167,6 +11167,32 @@ SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
   SDLoc dl(Op);
 
   switch (IntrinsicID) {
+  case Intrinsic::ppc_get_function_descriptor:
+    return Op.getOperand(1);
+  case Intrinsic::ppc_get_function_entry: {
+    SDValue Op1 = Op.getOperand(1);
+    if (auto *G = dyn_cast<GlobalAddressSDNode>(Op1)) {
+      const GlobalValue *GV = G->getGlobal();
+      assert(isFunctionGlobalAddress(GV));
+      assert(Subtarget.isAIXABI());
+      assert(!isa<GlobalIFunc>(GV) && "IFunc is not supported on AIX.");
+      const TargetMachine &TM = Subtarget.getTargetMachine();
+      const TargetLoweringObjectFile *TLOF = TM.getObjFileLowering();
+      MCSymbolXCOFF *S =
+          cast<MCSymbolXCOFF>(TLOF->getFunctionEntryPointSymbol(GV, TM));
+
+      MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
+      SDValue EntryPointSym = DAG.getMCSymbol(S, PtrVT);
+      return getTOCEntry(DAG, dl, EntryPointSym);
+    }
+    assert (Op1.getOpcode() == ISD::CopyFromReg);
+    SDLoc dl(Op);
+    EVT PtrVT = getPointerTy(DAG.getDataLayout());
+    SDValue Result = DAG.getLoad(
+            PtrVT, dl, DAG.getEntryNode(), Op1, MachinePointerInfo(), DAG.getDataLayout().getPointerABIAlignment(0),
+            MachineMemOperand::MODereferenceable);
+    return Result;
+  }
   case Intrinsic::thread_pointer:
     // Reads the thread pointer register, used for __builtin_thread_pointer.
     if (Subtarget.isPPC64())
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
index b5c6ac111dff0..d8e840912bc37 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -36,6 +36,7 @@
 #include "llvm/InitializePasses.h"
 #include "llvm/MC/TargetRegistry.h"
 #include "llvm/Pass.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
@@ -43,6 +44,7 @@
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/TargetParser/Triple.h"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/LowerIFunc.h"
 #include <cassert>
 #include <memory>
 #include <optional>
@@ -452,6 +454,15 @@ class PPCPassConfig : public TargetPassConfig {
 
 } // end anonymous namespace
 
+void PPCTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
+  if (getTargetTriple().isOSAIX())
+    PB.registerOptimizerLastEPCallback(
+        [](ModulePassManager &PM, OptimizationLevel Level, ThinOrFullLTOPhase Phase) {
+          if (Phase == ThinOrFullLTOPhase::None)
+            PM.addPass(LowerIFuncPass());
+        });
+}
+
 TargetPassConfig *PPCTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new PPCPassConfig(*this, PM);
 }
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.h b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
index cb02b446fadb3..608750f7a3f2b 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
@@ -54,6 +54,8 @@ class PPCTargetMachine final : public CodeGenTargetMachineImpl {
   // Pass Pipeline Configuration
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
 
+  void registerPassBuilderCallbacks(PassBuilder &PB) override;
+
   TargetTransformInfo getTargetTransformInfo(const Function &F) const override;
 
   TargetLoweringObjectFile *getObjFileLowering() const override {
diff --git a/llvm/lib/Transforms/Utils/LowerIFunc.cpp b/llvm/lib/Transforms/Utils/LowerIFunc.cpp
index 18ae0bbe2e731..991890aa56b83 100644
--- a/llvm/lib/Transforms/Utils/LowerIFunc.cpp
+++ b/llvm/lib/Transforms/Utils/LowerIFunc.cpp
@@ -13,6 +13,7 @@
 #include "llvm/Transforms/Utils/LowerIFunc.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
+#include "llvm/TargetParser/Triple.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
 
 using namespace llvm;
@@ -22,6 +23,10 @@ PreservedAnalyses LowerIFuncPass::run(Module &M, ModuleAnalysisManager &AM) {
   if (M.ifunc_empty())
     return PreservedAnalyses::all();
 
-  lowerGlobalIFuncUsersAsGlobalCtor(M, {});
+  Triple TargetTriple(M.getTargetTriple());
+  if (TargetTriple.isOSAIX())
+    lowerIFuncsOnAIX(M);
+  else
+    lowerGlobalIFuncUsersAsGlobalCtor(M, {});
   return PreservedAnalyses::none();
 }
diff --git a/llvm/lib/Transforms/Utils/ModuleUtils.cpp b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
index 596849ecab742..67da3e66cf16a 100644
--- a/llvm/lib/Transforms/Utils/ModuleUtils.cpp
+++ b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
@@ -16,6 +16,8 @@
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsPowerPC.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/Casting.h"
@@ -397,6 +399,85 @@ void llvm::embedBufferInModule(Module &M, MemoryBufferRef Buf,
   appendToCompilerUsed(M, GV);
 }
 
+void llvm::lowerIFuncsOnAIX(Module &M) {
+  for (GlobalIFunc &IFunc : make_early_inc_range(M.ifuncs())) {
+
+    // Let IFunc be:
+    //   @foo = ifunc rt-type (type1, type2), ptr @resolver
+    assert(isa<llvm::FunctionType>(IFunc.getValueType()));
+    FunctionType *FTy = cast<llvm::FunctionType>(IFunc.getValueType());
+
+    // Create the regular (non-ifunc) function:
+    //   define rt-type @foo(type1 %0, type2 %1)
+    Function *F = Function::Create(FTy, IFunc.getLinkage(), IFunc.getAddressSpace(), IFunc.getName(), &M);
+    LLVMContext &Ctx = F->getContext();
+
+    // entry:
+    BasicBlock *CurBlock = BasicBlock::Create(Ctx, "entry", F);
+    IRBuilder<> Builder(CurBlock);
+
+    PointerType *PtrTy = Builder.getPtrTy();
+    //   %DescPtr = call ptr @ppc_get_function_descriptor(ptr noundef @foo)
+    auto *DescPtr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_descriptor, {F}, {}, "desc.ptr");
+
+    //   %DesctAddr = getelementptr inbounds %struct.Desc_t, ptr %DescPtr, i32 0, i32 0
+    StructType *DescriptorType = StructType::get(PtrTy, PtrTy, PtrTy);
+    auto *DesctAddr = Builder.CreateStructGEP(DescriptorType, DescPtr, 0, "desc_t.addr");
+
+    //   %AddrInDesc = load ptr, ptr %DesctAddr, align 4
+    auto *AddrInDesc = Builder.CreateAlignedLoad(PtrTy, DesctAddr, DesctAddr->getPointerAlignment(M.getDataLayout()), "addr.in.desc");
+
+    //   %OriginalAddr = call ptr @ppc_get_function_entry(ptr noundef @foo) #3
+    auto *OriginalAddr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_entry, {F}, {}, "original.addr");
+
+    //   %cmp = icmp eq ptr %AddrInDesc, %OriginalAddr
+    auto *CMP = Builder.CreateICmpEQ(AddrInDesc, OriginalAddr);
+
+    //   br i1 %cmp, label %resolver, label %end
+    CurBlock = BasicBlock::Create(Ctx, "resolver", F);
+    BasicBlock *FinalBlock = BasicBlock::Create(Ctx, "end", F);
+    Builder.CreateCondBr(CMP, CurBlock, FinalBlock);
+
+    // Emit the 'then' (resolver) code:
+    Builder.SetInsertPoint(CurBlock);
+
+    // resolver:
+    //   %ResolvedFunc = call ptr @resolver()
+    //   %ResolvedAddr = call ptr @ppc_get_function_entry(ptr %ResolvedFunc)
+    //   %4 = ptrtoint ptr %ResolvedAddr to i32
+    //   store atomic i32 %4, ptr %desc_t.addr release, align 4
+    Function *ResolverFunc = IFunc.getResolverFunction();
+    auto *ResolvedFunc = Builder.CreateCall(ResolverFunc, ArrayRef<Value *>(), "resolved.func");
+    auto *ResolvedAddr = Builder.CreateIntrinsic(/*RetTy*/ PtrTy, Intrinsic::ppc_get_function_entry, {ResolvedFunc}, {}, "resolved.addr");
+    auto *PtrToInt = Builder.CreatePtrToInt(ResolvedAddr, Builder.getIntPtrTy(M.getDataLayout()));
+    // TODO fix alignment
+    Builder.CreateAlignedStore(PtrToInt, DesctAddr, MaybeAlign())->setAtomic(AtomicOrdering::Release);
+
+    //   br label %if.end
+    Builder.CreateBr(FinalBlock);
+
+    // Emit the continuation block for code after the if.
+    Builder.SetInsertPoint(FinalBlock);
+
+    //   %res = musttail call i32 %DescPtr(i32 noundef %a) #3
+    SmallVector<Value *, 10> Args(make_pointer_range(F->args()));
+    CallInst *Result = Builder.CreateCall(F->getFunctionType(), DescPtr, Args, "res");
+    //Result->setTailCallKind(CallInst::TCK_MustTail);
+
+    //   ret i32 %res
+    if (F->getReturnType()->isVoidTy())
+      Builder.CreateRetVoid();
+    else
+      Builder.CreateRet(Result);
+
+    // replace all uses of the ifunc with the newly created function
+    IFunc.replaceAllUsesWith(F);
+
+    std::string name = IFunc.getName().str();
+    IFunc.eraseFromParent();
+    F->setName(name);
+  }
+}
 bool llvm::lowerGlobalIFuncUsersAsGlobalCtor(
     Module &M, ArrayRef<GlobalIFunc *> FilteredIFuncsToLower) {
   SmallVector<GlobalIFunc *, 32> AllIFuncs;

github-actions · 2025-08-11T16:46:19Z

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff origin/main HEAD --extensions cpp,h,c -- compiler-rt/lib/builtins/ppc/init_ifuncs.c llvm/lib/Target/PowerPC/PPCPrepareIFuncsOnAIX.cpp clang/include/clang/Basic/TargetInfo.h clang/lib/Driver/ToolChains/AIX.cpp clang/test/CodeGen/attr-ifunc.c clang/test/CodeGen/attr-ifunc.cpp clang/test/CodeGen/ifunc.c clang/test/SemaCXX/ifunc-has-attribute.cpp llvm/include/llvm/CodeGen/AsmPrinter.h llvm/include/llvm/CodeGen/TargetLoweringObjectFileImpl.h llvm/include/llvm/Target/TargetLoweringObjectFile.h llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp llvm/lib/Target/PowerPC/PPC.h llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp llvm/lib/Target/PowerPC/PPCISelLowering.cpp llvm/lib/Target/PowerPC/PPCTargetMachine.cpp --diff_from_common_commit

⚠️
The reproduction instructions above might return results for more than one PR
in a stack if you are using a stacked PR workflow. You can limit the results by
changing origin/main to the base branch/commit you want to compare against.
⚠️

View the diff from clang-format here.

diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 5c4f9362c..480a510c6 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -3444,9 +3444,10 @@ static bool TOCRestoreNeededForCallToImplementation(const GlobalIFunc &GI) {
       }
       return Res;
     }
-    // @switch.table.resolve_foo = private unnamed_addr constant [3 x ptr] [ptr @foo_static, ptr @foo_hidden, ptr @foo_protected]
-    // %switch.gep = getelementptr inbounds nuw ptr, ptr @switch.table, i64 %2
-    // V = load ptr, ptr %switch.gep,
+    // @switch.table.resolve_foo = private unnamed_addr constant [3 x ptr] [ptr
+    // @foo_static, ptr @foo_hidden, ptr @foo_protected] %switch.gep =
+    // getelementptr inbounds nuw ptr, ptr @switch.table, i64 %2 V = load ptr,
+    // ptr %switch.gep,
     else if (auto *Op = getPointerOperand(I)) {
       while (isa<GEPOperator>(Op))
         Op = cast<GEPOperator>(Op)->getPointerOperand();
@@ -3587,7 +3588,8 @@ void PPCAIXAsmPrinter::emitGlobalIFunc(Module &M, const GlobalIFunc &GI) {
   if (TOCRestoreNeededForCallToImplementation(GI)) {
     reportFatalUsageError(
         "unimplemented: TOC register save/restore needed for ifunc \"" +
-        Twine(GI.getName()) + "\", because couldn't prove all candidates "
+        Twine(GI.getName()) +
+        "\", because couldn't prove all candidates "
         "are static or hidden/protected visibility definitions");
     return;
   }

w2yehia · 2025-08-15T02:55:28Z

I mistakenly thought ifunc can be forward declared with enough information to indicate it's an ifunc, so that all callsites of ifunc can be converted to indirect calls through the function descriptor (which is something that was planned as an immediate next step after this PR or included in it).
But today I found out it's not; an ifunc declaration looks like a regular declaration.
So the code in the function stub created during lowering became a focus point, and as it stands it requires some refinement.

I had some offline discussions with @RolandF77 and @mandlebug, and two solutions were proposed:

a constructor based approach, where the update of the descriptor is done in a constructor/initialization function. One init per ifunc or some aggregate constructor. The solution must not prevent dead functions from being garbage collected by the linker. One disadvantage is that: the resolvers of all (statically) live ifuncs are called at program startup, rather than individual resolvers lazily called upon function entry (which is the approach we tried here)
have a special function that calls the resolver and fixes up the descriptor, and tail calls the result.

In both approaches the original ifunc function becomes a wrapper that indirect calls the function pointer in the descriptor (no checks).

We would prefer a solution that doesn't require any input from the user (in the form of extra flags or user code changes (such as decorating ifunc declarations with some extra info)).

I will evaluate the two approaches and post an update here soon.

llvm/lib/Transforms/Utils/ModuleUtils.cpp

w2yehia

part 1 group codereview

clang/lib/Driver/ToolChains/AIX.cpp

clang/include/clang/Basic/TargetInfo.h

w2yehia · 2025-09-10T14:56:23Z

clang/test/CodeGen/attr-ifunc.c

@@ -4,6 +4,8 @@
 // RUN: %clang_cc1 -triple x86_64-apple-macosx -verify -emit-llvm-only %s
 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -verify -emit-llvm-only %s
 // RUN: %clang_cc1 -triple aarch64-pc-windows-msvcu -verify -emit-llvm-only %s
+// RUN: %clang_cc1 -triple powerpc64-ibm-aix-xcoff -verify -emit-llvm-only %s


shud have at least one 32-bit test here or in anothe LIT

since the test case only test whether the source code is compiled pass or not ? do it matter 32bit or 64bit ?

yes, maybe in this test it's not useful to test 32-bit, but we need some where to test that ifunc is supported in 32-bit AIX.

llvm/lib/Target/PowerPC/PPCPrepareIFuncsOnAIX.cpp

llvm/lib/IR/AsmWriter.cpp

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

clang/lib/Driver/ToolChains/AIX.cpp

diggerlin · 2025-09-12T15:23:18Z

clang/test/CodeGen/attr-ifunc.c

@@ -4,6 +4,8 @@
 // RUN: %clang_cc1 -triple x86_64-apple-macosx -verify -emit-llvm-only %s
 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -verify -emit-llvm-only %s
 // RUN: %clang_cc1 -triple aarch64-pc-windows-msvcu -verify -emit-llvm-only %s
+// RUN: %clang_cc1 -triple powerpc64-ibm-aix-xcoff -verify -emit-llvm-only %s


since the test case only test whether the source code is compiled pass or not ? do it matter 32bit or 64bit ?

compiler-rt/lib/builtins/ppc/init_ifuncs.c

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

In PR llvm#153049, we have a use case of attaching the !associated metadata to an ifunc. Since an ifunc is similar to a function declaration, it seems natural to allow metadata on ifuncs. Currently, the metadata API allows adding Metadata to llvm::Values, so the in-memory IR allows for metadata on ifuncs, but the IR reader/writer is not aware of that. Teach the IR parser and writer to support metadata on ifuncs, and update documentation.

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

hubert-reinterpretcast

When the target of the call is known to be an ifunc (e.g., when the definition is visible because of LTO), it would make sense to avoid going through the implementation stub and to just generate an indirect call sequence in-line from the call site. Such a call sequence is generated when -mlongcall is specified (but the IR realization of it seems to be tied to the caller and not the callee). LLVM has a "long-call" callee attribute for MIPS.

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

Teach the IR parser and writer to support metadata on ifuncs, and update documentation. In PR #153049, we have a use case of attaching the `!associated` metadata to an ifunc. Since an ifunc is similar to a function declaration, it seems natural to allow metadata on ifuncs. Currently, the metadata API allows adding Metadata to llvm::GlobalObject, so the in-memory IR allows for metadata on ifuncs, but the IR reader/writer is not aware of that. --------- Co-authored-by: Wael Yehia <[email protected]>

…ion -bdbg:namedsects:ss by default on AIX

…art/stop symbol references when user code does not use ifuncs

also pass -bdbg:namedsects:ss if AIX version is unknown

Load R11 in the ifunc stub

…ntation

…ocal()

mandlebug

Partway through my review but though I'd leave some of the initial comments.

clang/include/clang/Basic/TargetInfo.h

clang/lib/Driver/ToolChains/AIX.cpp

compiler-rt/lib/builtins/ppc/init_ifuncs.c

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

mandlebug · 2025-09-25T14:35:04Z

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

    cl::desc("Enable Passing SSP Canary info in Trackback on AIX"), cl::Hidden);

+static cl::list<std::string> IFuncLocal(
+    "ifunc-local",


I don't recall where we ended up in the discussion of this option. I'm not a fan of having a safelist option that takes a list of strings, however I think we came to the conclusion this might be needed because having a compiler only implementation can't sufficiently verify the resolver is always safe?

We didn't settle on whether to keep it or not.

I spoke with Sean offline, and he suggested to make it a user error to have resolvers returning non-local functions, and to change the walker in function TOCRestoreNeededForCallToImplementation to flag any violation it can find.

@mandlebug please see latest version of TOCRestoreNeededForCallToImplementation.
I changed the walker to return a tristate: IsLocal (all are provably local), IsPreemptible (at least one is provably preemptible), and Unknown (a mix of provablty local and unknown structure).
I added a false-by-default flag -mllvm -ifunc-local-if-proven to control the bias. By default, the bias is local unless proven preemptible. I thought some users might not be comfortable with this as it might lead to silent bad behavior, so they can opt-out by using that flag. The flag is boolean, not a list of functions.

Once we agree that this is an acceptable approach, I'll add a note under the ifunc attribute documentation.

mandlebug

Some minor comments so far.

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp

llvm/lib/Target/PowerPC/PPCPrepareIFuncsOnAIX.cpp

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp

llvmbot added clang Clang issues not falling into any other category backend:PowerPC clang:frontend Language frontend issues, e.g. anything involving "Sema" llvm:ir llvm:transforms labels Aug 11, 2025

w2yehia force-pushed the ifunc branch from a6c8637 to c294b43 Compare August 11, 2025 19:01

MaskRay reviewed Aug 15, 2025

View reviewed changes

llvm/lib/Transforms/Utils/ModuleUtils.cpp Outdated Show resolved Hide resolved

w2yehia force-pushed the ifunc branch from d77843c to b85dd0b Compare September 9, 2025 23:23

llvmbot added llvm:codegen compiler-rt clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' compiler-rt:builtins labels Sep 9, 2025

w2yehia commented Sep 10, 2025

View reviewed changes

w2yehia requested review from daltenty, hubert-reinterpretcast and mandlebug September 12, 2025 14:25

diggerlin reviewed Sep 12, 2025

View reviewed changes

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp Outdated Show resolved Hide resolved

w2yehia mentioned this pull request Sep 15, 2025

[IR] enable attaching metadata on ifuncs #158732

Merged

w2yehia force-pushed the ifunc branch from 12c4d26 to 26eff05 Compare September 16, 2025 00:53

hubert-reinterpretcast reviewed Sep 16, 2025

View reviewed changes

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp Outdated Show resolved Hide resolved

hubert-reinterpretcast reviewed Sep 16, 2025

View reviewed changes

llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp Outdated Show resolved Hide resolved

hubert-reinterpretcast reviewed Sep 16, 2025

View reviewed changes

w2yehia force-pushed the ifunc branch from 26eff05 to 629fe41 Compare September 18, 2025 03:05

w2yehia commented Sep 18, 2025

View reviewed changes

Wael Yehia added 9 commits September 21, 2025 01:34

accept ifunc attribute on AIX

c562d40

enable tests

7defcf4

Add ppc/init_ifuncs.c to the builtins library and pass the linker opt…

916b94e

…ion -bdbg:namedsects:ss by default on AIX

create a zero-length entry in the ifunc_sec section to satisfy the st…

5a1b3ce

…art/stop symbol references when user code does not use ifuncs

use -bdbg:namedsects:ss on AIX 7.2 and above

36f1e88

also pass -bdbg:namedsects:ss if AIX version is unknown

Address code review comments

27da809

Load R11 in the ifunc stub

comment changes and logic updates in TOCRestoreNeededForCallToImpleme…

896b543

…ntation

refactor TOCRestoreNeeded

13868e8

simplify IsLocalFunc by using isStrongDefinitionForLinker() && isDSOL…

204ea53

…ocal()

w2yehia force-pushed the ifunc branch from 9b377ca to 204ea53 Compare September 23, 2025 02:25

improve switch table detection

8f81cc6

remove assert in transformCallee()

4956ea2

mandlebug reviewed Sep 25, 2025

View reviewed changes

w2yehia mentioned this pull request Sep 26, 2025

XCOFF associated metadata #159096

Open

Wael Yehia added 3 commits September 30, 2025 13:05

init_ifuncs.c: copy entire descriptor

28f128c

code review

62b6e6f

refactor TOCRestoreNeededForCallToImplementation based on code review

9b7b796

mandlebug reviewed Oct 3, 2025

View reviewed changes

Wael Yehia added 5 commits October 7, 2025 23:12

Sean's second code review

4c7462d

ifunc test 1

d453395

more ifunc tests

e55fb54

update section name and refactor init_ifuncs.c

40856e1

improve error message

a25cbf1

w2yehia force-pushed the ifunc branch from bd4bf85 to c8094b6 Compare December 2, 2025 20:06

Wael Yehia added 2 commits December 2, 2025 20:14

support large code model

c8094b6

add comments about the enum IsLocal

6ba9e24

[AIX] Implement the ifunc attribute. #153049

Are you sure you want to change the base?

[AIX] Implement the ifunc attribute. #153049

Uh oh!

Conversation

w2yehia commented Aug 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Aug 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Aug 11, 2025

Uh oh!

github-actions bot commented Aug 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

w2yehia commented Aug 15, 2025

Uh oh!

Uh oh!

w2yehia left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

w2yehia Sep 10, 2025

Choose a reason for hiding this comment

Uh oh!

diggerlin Sep 12, 2025

Choose a reason for hiding this comment

Uh oh!

w2yehia Sep 12, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

diggerlin Sep 12, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hubert-reinterpretcast left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mandlebug left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mandlebug Sep 25, 2025

Choose a reason for hiding this comment

Uh oh!

w2yehia Sep 30, 2025

Choose a reason for hiding this comment

Uh oh!

w2yehia Sep 30, 2025

Choose a reason for hiding this comment

Uh oh!

w2yehia Oct 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mandlebug left a comment

Choose a reason for hiding this comment

w2yehia commented Aug 11, 2025 •

edited

Loading

llvmbot commented Aug 11, 2025 •

edited

Loading

github-actions bot commented Aug 11, 2025 •

edited

Loading

w2yehia Oct 1, 2025 •

edited

Loading